We Must Return to the Office to Use AI in Person

Training tries to produce something that scores highly in training evaluations. With one data point, the evaluation is solely how closely the model output resembles the single input text.

Let's say you do that, and the training text is 58,100 tokens long. Let's say you ask the model to produce 58,101 tokens. Will it "reproduce [the] text verbatim"? No, it can't, because of the dissimilar requested length. Something "new" will come out.

It's also entirely possible that no matter how long you train, the model never converges on generating exactly the same output as its training data - you could end up with an average loss value of 0.001 instead of 0.0. It's not a perfectly deterministic process.

You're correct in principle, but in reality even with limited training data real-world models produce something that isn't exactly their training, especially when sampled stochastically. They're biased toward their training data, not forced to it.

In a physical space, humans have built-in advantages over bots. Operating exclusively or primarily over IP levels the playing field, nullifying our advantages of expressive nonverbal communication, and makes 100% of our output accessible to agents.

And it minimizes our contact with each other. I know you can DM people on Slack to gossip about the boss, but do you? Figuring everything is logged, you'd have to create and nurture alternative private channels like texting, and doing that at scale carries at least some implication of subterfuge that limits participation, in a way that chatting around a table at the sandwich shop just doesn't.

All of us really do appreciate the savings on gas money and (in theory) the time savings, though I bet most people have given over some of that time savings to work tasks. I'm not saying it doesn't have benefits. But I don't think the companies still embracing remote work are doing it out of philanthropy, and I don't think it's necessarily good for us workers.